gdbserver with reversed arguments goes into an infinite loop

Denis PILAT denis.pilat@st.com
Wed Dec 13 15:46:00 GMT 2006


Daniel Jacobowitz wrote:
> On Wed, Dec 13, 2006 at 10:58:14AM +0100, Denis PILAT wrote:
>> I've found that if you revert the argument of gdbserver, means writing 
>> the program's name before the COMM argument, it goes into an infinite 
>> loop, and as the CTRL+C does not work, you have to kill the process from 
>> an other shell.
>>
>> In gdbserver/server.c, the loop in question does the remote_open on the 
>> wrong passed argument (argv[1]) which unfortunately is the binary file 
>> you'd expect to open so remote_open does not exit on error.
>>
>> I think either we could check that we pass correct argument before using 
>> start_inferior(), this is executing before the loop. The bellow patch is 
>> in that sense.
>>
>> Or we find a way to exit the loop by adding a test in it. May be by 
>> adding something in remote_open to let it fail.
>
> I doubt it's in an infinite loop.  It's probably sleeping, "waiting"
> for a connection.  We ought to allow C-c when no debugger is
> connected yet.  It'd be nice if remote_open wouldn't open ordinary
> files, too.
>
Yes it's waiting for a connection but as arguments are wrongs it can 
wait for long ...

You're right, preventing remote_open from opening ordinary files would 
exit the loop.
But may be it would be better to open only character device (S_ISCHR 
macro) than excluding ordinary files (S_ISREG macro). It's up to you !
Here is a patch where I made a test to open only character device. If 
you right with that solution I'll propose a patch with a ChangeLog and 
so on.
I took this opportunity to remove a warning on a strncpy() usage.

I'm wondering about the compilation of this code under windows. I never 
compiled a gdbserver on windows, is there any gdbserver hosted under 
windows ?

-- 
Denis

Index: remote-utils.c
===================================================================
--- remote-utils.c      (revision 544)
+++ remote-utils.c      (working copy)
@@ -36,6 +36,7 @@
 #include <sys/time.h>
 #include <unistd.h>
 #include <arpa/inet.h>
+#include <sys/stat.h>

 #ifndef HAVE_SOCKLEN_T
 typedef int socklen_t;
@@ -68,10 +69,24 @@ void
 remote_open (char *name)
 {
   int save_fcntl_flags;
-
-  if (!strchr (name, ':'))
+  char *port_str;
+
+  port_str = strchr (name, ':');
+
+  /* if name is not of kind "HOST:PORT" it must be tty device.  */
+  if (!port_str)
     {
-      remote_desc = open (name, O_RDWR);
+      struct stat status;
+      int stat_result;
+      remote_desc = -1;
+
+      /* Open only character device.  */
+      stat_result = stat(name, &status);
+      if (!stat_result && S_ISCHR(status.st_mode))
+      {
+        remote_desc = open (name, O_RDWR);
+      }
+
       if (remote_desc < 0)
        perror_with_name ("Could not open remote device");

@@ -123,14 +138,11 @@ remote_open (char *name)
     }
   else
     {
-      char *port_str;
       int port;
       struct sockaddr_in sockaddr;
       socklen_t tmp;
       int tmp_desc;

-      port_str = strchr (name, ':');
-
       port = atoi (port_str + 1);

       tmp_desc = socket (PF_INET, SOCK_STREAM, 0);
@@ -650,7 +662,7 @@ prepare_resume_reply (char *buf, char st
          CORE_ADDR addr;
          int i;

-         strncpy (buf, "watch:", 6);
+         buf = strncpy (buf, "watch:", 6);
          buf += 6;

          addr = (*the_target->stopped_data_address) ();




More information about the Gdb-patches mailing list